danilsko / navalny

navalny stylometry

navalny

Efforts to reproduce "Navalny's texts from jail are not written by Navalny" claim and stylometric results.

Data

We gathered some stuff manually, but also a sizeable share of our data comes from: https://github.com/norpadon/navalny (many thanks!)
See 'corpus' folder and other folders in this repo

Research

tl;dr (so far)

////

So far, two strong candidates for “writings-from-jail” style emerged: 1) Navalny himself and 2) Maria Pevchikh. Likely it’s a mixed corpus.

////

1. Navalny_free vs Navalny_jail samples show stable heterogeneity: there is similarity between them, but the difference is obvious, too.
2. When verified against potential candidates, Navalny_jail exhibits mixed signal, both Navalny_free and Maria Pevchikh show stylistic affinities with jail samples.
3. Overall, between-author relationships are muddied (e.g Zhdanov and Volkov show certain level of similarity), which can be a trace of editor/ghostwriter.
4. Corpus needs more attention: 1) I can't rule out some systematic bias: what we might be seeing is a 'platform signal', not authorship. 2) We need more Y. Navalnaya's samples; we see she is similar to Pevchikh and Navalny, but not enough data to do proper verification. 3) Something is off about Sobol samples that come from large telegram data (duplicates?).

Full research:

Notebook for R and stylo (renders on GH)

Misc

Check also this attempt on much smaller corpus.

navalny's Issues

add Kira Yarmysh sample

Since Kira Yarmysh is an FBK's and Navalny's press-secretary, it makes sense to check her texts' similarity to Navalny's posts (both before and in prison).